Saltar al contenido principal
Model logo

API de Gemini Omni Flash 1.2

  • Texto a vídeo
  • Imagen a vídeo
  • Referencia a vídeo
  • Edición de video

Gemini Omni Flash 1.2 forma parte de la familia de modelos GeminiVer todos los modelos Gemini

Modelo

Sube imágenes/vídeos arriba y luego escribe @ en tu prompt para referenciarlos

Toca para subirPNG, JPG, WEBP, GIF · Máx. 100 MB
Toca para subirPNG, JPG, WEBP, GIF · Máx. 100 MB
Toca para subirPNG, JPG, WEBP, GIF · Máx. 100 MB
Salida

Tu video generado aparecerá aquí

Actualizado

What is the Gemini Omni Flash 1.2 API?

The Gemini Omni Flash 1.2 API is live on UnificAlly as google/gemini-omni-flash-1.2-video, from $0.019 per second of video. Gemini Omni Flash 1.2 is Google's multimodal video model, released November 5, 2026 as the next step after Gemini Omni Flash 1.1. Text, still images, named characters, and a source video all go into the model context, and the output is video with audio. Google orders the name as Gemini Omni 1.2 Flash; on UnificAlly it runs as google/gemini-omni-flash-1.2-video for generation, google/gemini-omni-flash-1.2-video-edit for editing, and google/gemini-omni-flash-1.2-video-extend for continuing a finished video.

The update is about control. Clips can now be any whole number of seconds from 3 to 15, so a 7 second product shot or a 13 second scene no longer has to be padded or cut to fit a fixed length. References hold better: a character or setting you upload keeps its look across the clip instead of drifting after a cut. Jump cuts are easier to direct from the prompt, and the picture is sharper overall.

The request body is the same as 1.1. Existing integrations move over by swapping the model string; the wider duration range and the longer edit window are new options on top.

Gemini Omni Flash 1.2 API specs

Model IDgoogle/gemini-omni-flash-1.2-video
Edit model IDgoogle/gemini-omni-flash-1.2-video-edit
EndpointPOST https://api.unifically.com/v1/tasks
Request typesText-to-video, first and last frame, reference-to-video, video edit
Duration3 to 15 seconds, any whole second. Up to 10 seconds with a first and last frame
Resolution360p, 720p, 1080p, 4K. 1080p and 4K are upscaled in the same task
Aspect ratio16:9, 9:16, auto
AudioVideo with audio, up to 5 Google voice presets
ReferencesUp to 7 images and 3 characters. Edit takes 1 source video, up to 5 images and 3 characters
Rate at 360p, 3 to 13 seconds$0.019/sec
Rate at 360p, 14 to 15 seconds$0.019/sec
Rate at 720p, 3 to 13 seconds$0.0428/sec
Rate at 720p, 14 to 15 seconds$0.0428/sec
Rate at 1080p$0.0475/sec
Rate at 4K, 5 to 15 secondscurrent rate
Rate at 4K, 3 to 4 secondscurrent rate

Billing is per second of output, with no subscription. Edits are billed at the same rates, by the length of the edited window. The pricing page shows the live rates for the Gemini Omni Flash 1.2 API.

How to use the Gemini Omni Flash 1.2 API

  1. Create an API key on the API keys page.
  2. Send a POST request to /v1/tasks with the model ID and an input object.
  3. Read the task_id from the response, then check the task until it is finished, or pass a callback URL and wait for the result.
curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/gemini-omni-flash-1.2-video",
    "input": {
      "prompt": "@Character1 walks into a sunlit bakery and picks up a croissant. Cut to a close-up of her first bite.",
      "reference_characters": [
        {
          "image_urls": [
            "CHARACTER_IMAGE_URL"
          ],
          "name": "Mia"
        }
      ],
      "duration": 7,
      "resolution": "720p",
      "aspect_ratio": "9:16"
    }
  }'

The response returns the task ID straight away:

{
  "code": 200,
  "success": true,
  "data": {
    "task_id": "abc123def456",
    "status": "processing"
  }
}

Poll the task, or set a callback_url on the request to receive the finished result. The finished task carries the video URL.

curl https://api.unifically.com/v1/tasks/{task_id} \
  -H "Authorization: Bearer YOUR_API_KEY"

Gemini Omni Flash 1.2 API parameters

ParameterTypeDefaultWhat it does
promptstringnoneMain video prompt. Required. Use @ImageN or @CharacterN to point at specific references
seedintegerrandom0 to 2147483647. Reuse a seed for reproducible results
durationinteger4Clip length in whole seconds, 3 to 15. First and last frame mode allows up to 10
resolutionstring720p360p, 720p, 1080p, or 4k. 360p and 720p generate natively. 1080p and 4k are upscaled after generation in the same task
aspect_ratiostring16:916:9, 9:16, or auto. auto follows the shape of start_image_url or the first reference image. Text-only requests return 16:9
start_image_urlstringnoneImage used as the first frame. Cannot be combined with reference_image_urls or reference_characters
end_image_urlstringnoneImage used as the last frame. Needs start_image_url
reference_image_urlsstring[]noneUp to 7 reference images. Cannot be combined with the frame fields
reference_charactersarraynoneUp to 3 characters, each an object with 1 to 10 image_urls. Name them in the prompt as @Character1, @Character2, @Character3
voicestring or string[]noneGoogle voice preset ID, or an array of up to 5. Needs at least one reference image or character
allow_audio_filteredbooleanfalseWhen true, the video is returned without audio instead of failing when the audio track is filtered
nsfw_checkbooleanfalseChecks the prompt and input images for NSFW content before generating

The Gemini Omni Flash 1.2 API documentation has the full reference. The edit model has its own page at docs.unifically.com/models/video/google/gemini-omni-flash-1.2-video-edit, and the extend model at docs.unifically.com/models/video/google/gemini-omni-flash-1.2-video-extend.

First and last frame request

Provide both frames and the clip animates from start_image_url to end_image_url.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/gemini-omni-flash-1.2-video",
    "input": {
      "prompt": "The flower blooms in a smooth timelapse",
      "start_image_url": "https://example.com/bud.jpg",
      "end_image_url": "https://example.com/bloom.jpg",
      "aspect_ratio": "16:9",
      "duration": 6
    }
  }'

Video edit request

The edit model takes exactly one source video in reference_video_urls. start_frame and end_frame pick the window to change, up to 15 seconds.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/gemini-omni-flash-1.2-video-edit",
    "input": {
      "reference_video_urls": [
        "https://example.com/source-video.mp4"
      ],
      "prompt": "Make @Video1 look like it was filmed underwater.",
      "start_frame": 0,
      "end_frame": 288,
      "resolution": "720p"
    }
  }'

What's new in Gemini Omni Flash 1.2

Any length from 3 to 15 seconds

duration takes any whole number of seconds from 3 to 15, up from the fixed 4, 6, 8, or 10 seconds in 1.1. First-and-last-frame mode allows up to 10 seconds.

Stronger reference adherence

Characters and scenes from reference_characters and reference_image_urls carry through more faithfully, so faces, outfits, products, and settings stay recognizable for the whole clip.

Better control over jump cuts

Describe a cut in the prompt and the model follows it more closely, which makes short multi-shot sequences practical inside one generation.

Sharper visuals

Output is sharper than 1.1 at the native 360p and 720p resolutions. 1080p and 4K are upscaled after generation in the same task.

Extend a finished video

The extend model continues a completed generation or an earlier extension by 3 to 10 seconds and returns the whole video in one file. Only the new seconds are billed.

Longer edits, billed by length

The edit model now changes a window of up to 15 seconds from one source video. Edits are priced like generations, by the length of the edited window, with a 3 second minimum.

What Gemini Omni Flash 1.2 is best for

  • Character-led short videos

    Upload up to three named characters and call them out as @Character1, @Character2 in the prompt when the same people need to appear across shots.

  • Exact-length clips

    Ads, social cutdowns, and title cards that need a specific length, from 3 to 15 seconds, without trimming afterwards.

  • Multi-shot sequences

    Short scenes with deliberate cuts between angles or moments, directed from one prompt.

  • Edits of an existing video

    Restyle a clip, change the setting, or replace a subject across up to 15 seconds of one source video.

  • Scenes built step by step

    Generate a clip, then extend it 3 to 10 seconds at a time with a new prompt for each part, steering the scene as it grows.

  • Cheap prompt drafts

    Test direction at native 360p, then rerun the version you like at 720p, 1080p, or 4K with the same seed.

  • Voice-guided scenes

    Attach a reference and pick up to five voice presets when the video needs spoken lines from specific characters.

Gemini Omni Flash 1.2 use cases

Build a 12 second product ad from a packshot and a lifestyle still, with a cut from the hero shot to the product in use. Keep a recurring mascot on-model across a series of short videos by uploading the same character reference each time. Turn a start frame and an end frame into a 6 second transition between two layouts. Take a 30 second creator video and rework a 15 second stretch of it into a new setting while the rest stays untouched. Draft five prompt variations at 360p, pick one, and render it at 4K. Generate an 8 second opening shot, then extend it twice to follow the action into the next two beats.

Extend a video

Extend continues a finished Gemini Omni Flash 1.2 video. Send the task_id of a completed generation or an earlier extension to google/gemini-omni-flash-1.2-video-extend, describe what happens next in prompt, and set duration to the number of seconds to add, any whole number from 3 to 10.

The result is the whole video: the source clip with the new part joined on, in one file. The new part keeps the source clip's resolution and aspect ratio, so there is no resolution or aspect_ratio field, and the source has to be a 360p or 720p video. You can extend the result again to build a longer scene step by step.

Billing counts only the new seconds, at the source clip's resolution: $0.019/sec for a 360p source and $0.0428/sec for a 720p source.

Extend takes up to 3 reference images, up to 2 characters, and up to 3 voice presets for the new part. One limit to plan around: a video that was made with character references cannot be extended. Full parameters are in the extend docs.

Gemini Omni Flash 1.2 limitations

Generation runs prompt, frame, and reference modes; frames and references cannot be combined in one request. Extend adds 3 to 10 seconds per call to a 360p or 720p video and does not work on a video made with character references.

First-and-last-frame mode is capped at 10 seconds, and the end frame needs a start frame. Generation takes up to 7 reference images and 3 characters; edit takes exactly one source video plus up to 5 reference images and 3 characters. In generation, aspect_ratio is 16:9, 9:16, or auto, which follows the start image or the first reference image and gives 16:9 for a text-only request.

360p and 720p are the native resolutions. 1080p and 4K are upscaled after generation, not generated natively at those sizes.

Audio input stays narrow: voice presets and per-character voice fields are the supported path, and the model does not take an arbitrary uploaded audio track. Speech, captions, brand marks, and people-centric edits still need a review pass. Generated videos carry Google's SynthID watermark.

Gemini Omni Flash 1.2 vs Gemini Omni Flash 1.1

For API users the differences are length and control. 1.1 generates 4, 6, 8, or 10 second clips; 1.2 takes any whole second from 3 to 15. 1.1 edits a window of up to 10 seconds at a flat price per resolution; 1.2 edits up to 15 seconds and bills by the edited length, so short edits cost less. 1.2 also follows references and prompted cuts more closely and renders sharper detail. Both share the same request body, and 1.1 is on its way out, so new work belongs on 1.2.

When to use Gemini Omni Flash 1.2

Use Gemini Omni Flash 1.2 when a video job needs more than a prompt: named characters that must stay consistent, a specific clip length, a cut you want to direct, a first and last frame, voice presets, a targeted edit of an existing video, or a finished clip that needs to run longer.

Preguntas frecuentes

Preguntas frecuentes sobre Gemini Omni Flash 1.2

Gemini Omni Flash 1.2 is Google's multimodal video model, released November 5, 2026 as the successor to Gemini Omni Flash 1.1. It takes text, images, character references, and a source video, and returns video with audio. Google writes the name as Gemini Omni 1.2 Flash.

Get access by registering and creating a key on the API keys page; it rides in the Authorization: Bearer <key> header to https://api.unifically.com/v1/tasks. One key drives all three Omni Flash 1.2 IDs (generate, edit and extend), and video is paid per second of output from a prepaid balance.

For a new clip, send google/gemini-omni-flash-1.2-video a prompt plus optional duration (any whole number from 3 to 15 seconds), resolution, aspect_ratio, start and end frame images or reference_characters; edits go to google/gemini-omni-flash-1.2-video-edit and continuations to google/gemini-omni-flash-1.2-video-extend. Each call returns a task_id, and the finished task at /v1/tasks/{task_id}, or your callback_url, carries the video URL.

Four things. Clips can be any whole number of seconds from 3 to 15 instead of 4, 6, 8, or 10. Characters and scenes from your references carry through more faithfully. Jump cuts inside a clip are easier to control from the prompt. Visuals are sharper. On the edit side, the window you change can now be up to 15 seconds long.

Three. Use google/gemini-omni-flash-1.2-video for text-to-video, first-and-last-frame, and reference-to-video generation. Use google/gemini-omni-flash-1.2-video-edit when you already have one source video and want to change it with a prompt. Use google/gemini-omni-flash-1.2-video-extend to continue a finished Omni Flash 1.2 video by 3 to 10 seconds. Generate and edit take the same request body as their 1.1 counterparts, so moving over is a change of the model string.

Generation takes any whole number of seconds from 3 to 15, set with duration. First-and-last-frame mode (start_image_url plus end_image_url) allows up to 10 seconds. Edit mode changes a window of up to 15 seconds from one source video, picked with start_frame and end_frame. Extend adds 3 to 10 seconds per call to a finished video, and you can extend the result again.

Yes. Send the task_id of a finished Omni Flash 1.2 generation or earlier extension to google/gemini-omni-flash-1.2-video-extend with a prompt for what happens next and a duration from 3 to 10 seconds. The result is the whole video, source plus the new part, and only the new seconds are billed, at the source clip's resolution. A video made with character references cannot be extended.

Edits are billed the same way as generations, by the length of the edited window and the output resolution. The window is rounded up to whole seconds with a 3 second minimum, so editing 7.2 seconds of a clip costs the same as generating an 8 second clip at that resolution. Rates per second start at $0.019/sec for 360p, $0.0428/sec for 720p, and $0.0475/sec for 1080p. The pricing page shows the live table.

360p and 720p are generated natively. 1080p and 4K are upscaled after generation inside the same task, so you get the finished file from one request. 360p is the cheap draft tier for testing prompts before a final 720p, 1080p, or 4K run.

Yes. Output is video with audio. Generation accepts up to five Google voice presets when at least one image or character reference is attached, and edit mode supports per-character voice fields. Set allow_audio_filtered to true if you would rather get a silent video than a failed task when the audio track is filtered.

Yes, for now. The Gemini Omni Flash 1.1 page and endpoints stay live, but 1.1 is on its way out, so new projects should start on 1.2. Existing 1.1 requests move over by changing the model string.